A Parallel Image Fusion Algorithm Based on Wavelet Packet

Yinglei Cheng1,2, Ying Li2, Rongchun Zhao2

2010/01/07

Outline
ABSTRACT

1. Introduction
2. Wavelet Packet Fusion Algorithm

Estimation of Computation Cost 3. Parallel Fusion Algorithm 4. Analysis of Parallel Algorithm Property 5. Conclusion

ABSTRACT
The wavelet packet provides an accurate method

for image fusion.

we achieve the algorithm on the MPI (Message

Passing Interface) parallel computing platform.

1. Introduction
Recently, image fusion has become one of the

focuses in image processing field.

Especially for fusion method based on the

wavelet packet transform, with the increasing of the image size and algorithm decomposition level, its computing cost will increase even more.

2. Wavelet Packet Based Serial Fusion Algorithm Estimation of Computation Cost

the first step of multi-spectral TM image and SAR

multi-resolution image serial fusion is the preprocessing of the whole TM image.

the image is decomposed into 4 sub-images: LL,

HH, LH and HL, and then each sub-image is decomposed again into 4 sub-sub-images.
With the increase of the image size, the

decomposition levels increase in a nonlinear way.

3. Parallel Fusion Algorithm

3.1. The Theory of Parallel Wavelet Packet Algorithm

The basic idea is to separate the image with the

size of M * N into the machine P parts that contain M *N / P pixels in each one. Then every computer only needs to compute M*N / P data.

3.2. Parallelization of the Wavelet Packet Algorithm

The Wavelet Packet transform is the

improvement of wavelet transform.

For an image with the size M *N , the value of its

every point is x k1,k2 .

3.3 . The Achievement of the Parallel Fusion Algorithm

The program is compiled with language VC++. All

steps of the algorithm are as follows:

(1) Pre-processing (2) Data partition (3) Locally output range computation (4) Locally fusion and reconstruction computation (5) Post-processing

4. Analysis of Parallel Algorithm Property

the speed of serial algorithm. It is defined as

follows:

The parallel efficiency is defined as follows:

4.1. Linear Weighted Value Comparison

TM multi-spectrum image and SAR high space-

resolution image are fused according to the weighted fusion rule in this paper. The weighted computation formula is:

4.2. Analysis of Parallel Efficiency

It is put forward in this paper that the parallel

algorithm can fuse SAR image and TM multispectrum image effectively.

5. Conclusion
first proposed fusion algorithm of multi-sensor

image bases on the wavelet packet transform, then analyzed the potential parallel capability.
The wavelet packet transform image fusion is a

new kind of fusion technique with excellent performance.

this paper discusses its implementation in a rapid

and real time way, it is of certain significance to the popularization and application of the wavelet packet based image fusion.

THANKS